Thanks for the detailed and informative issue ! This is indeed a critical one, whose fix is on the way.

To the interested audience, this is how git compares tree entries:

int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
{
    size_t min_len = (len1 < len2) ? len1 : len2;
    int cmp = memcmp(name1, name2, min_len);
    if (cmp)
        return cmp;
    if (len1 < len2)
        return -1;
    if (len1 > len2)
        return 1;
    return 0;
}

Now git-python does it similarly. It's worth noting that the most difficult was to make it work in Py3 as well as in Py2, which contributes to most of the added code.

I am hoping to hear from you if it does indeed fix the issue on your end :).
Thanks again.

Whoa, hadn't thought this would be this complicated. Big thanks for the quick response!

I will test if the bug is fixed for our repositories, but I'm pretty certain we will have no more issues.

Sadly I still get errors. I took one tree which produces errors and used "git show" to see what is stored in object store. The attached files correspond to the four different ways I used to create the tree:

• git.txt - Original git version, this is correct
• gitpython.txt - Generated by gitpython directly downloaded form git (including d82a6c5), this should be correct
• gitpython_reversed.txt - same as above, but I switched len_b - len_a to len_a - len_b (just tried this, could have worked)
• gitpython_old.txt - Original version of gitpython I used when first discovering the bug

git seems to do some additional magic? Or am I missing something?

Thanks a lot ! With the new, improved dataset, I am sure there will soon be an actual fix.
This makes me curious what's so special about how git does it, after all I was looking at git's source code and thought I now did it by the book.
Maybe it's the sorting algorithm - git uses quick sort, and I am not quite sure what python's List.sort is actually doing.

Please have a look at the two commits related to the issue to see if this can indeed solve the problem.

Long story short: The API used here shouldn't have existed in the first place, as it's not possible to sort individual trees correctly. Instead, please resort to modifying index entries and use IndexFile.write_tree() to write out the tree objects for use in a commit.

Of course, if this still is not working or creates invalid repositories, please let me know right away so I can have another look.

Thanks for your effort so far, I found the input very valuable.